Plato's Cave Algorithm: Inferring Functional Signaling Networks from Early Gene Expression Shadows
Figure 3
Weight Matrix Derivation – choosing good predictors of signaling activity.
A. The expression levels (in arbitrary units) of four genes under five perturbations (red bars), and their levels without any perturbation (green bars); B. The Z-values for each gene under each perturbation. High Z-values are obtained for genes with statistically significant change (red bars), and low Z-values are obtained when the change can be attributed to noise (blue bars); C. The information score of each gene. When the change in gene activity is specific to one perturbation the information score is high (red bars), and otherwise it is low (blue bars); D. The final weight attributed to each gene for each perturbation. A gene can only get a high weight (red bars) if it has both a high Z-value and a high information score. The values used for the gene expression are not drawn from any experiment and were generated merely to illustrate the methodology.